#include <stdio.h>
#include <stdlib.h>
#include <stdint.h>
#include <string.h>
#include <math.h>
#include <limits.h>
#include <time.h>

#include "nmglobal.h"
#include "randnum.h"
#include "emfloat.h"


/*****************************
** FLOATING-POINT EMULATION **
*****************************/

static clock_t DoEmFloatIteration(InternalFPF *abase, InternalFPF *bbase, InternalFPF *cbase,
        unsigned long arraysize, unsigned long loops);
static void SetupCPUEmFloatArrays(InternalFPF *abase, InternalFPF *bbase, unsigned long arraysize);

/**************
** DoEmFloat **
***************
** Perform the floating-point emulation routines portion of the
** CPU benchmark.  Returns the operations per second.
*/
void
DoEmFloat(void)
{
        const char*     errorcontext = "CPU:Floating Emulation";
        EmFloatStruct*  locemfloatstruct = &global_emfloatstruct;
        InternalFPF*    abase = NULL;
        InternalFPF*    bbase = NULL;
        InternalFPF*    cbase = NULL;
        clock_t         total_time = 0;
        int             iterations = 0;
        unsigned long   loops = 1;

        abase = malloc(locemfloatstruct->arraysize * sizeof(InternalFPF));
        if (!abase) {
                printf("ERROR CONDITION\nContext: %s\n", errorcontext);
                exit(1);
        }
                

        bbase = malloc(locemfloatstruct->arraysize * sizeof(InternalFPF));
        if (!bbase) {
                printf("ERROR CONDITION\nContext: %s\n", errorcontext);
                free(abase);
                exit(1);
        }

        cbase = malloc(locemfloatstruct->arraysize * sizeof(InternalFPF));
        if (!cbase) {
                printf("ERROR CONDITION\nContext: %s\n", errorcontext);
                free(abase);
                free(bbase);
                exit(1);
        }
        
        SetupCPUEmFloatArrays(abase, bbase, locemfloatstruct->arraysize); /* FIXME: ugly */

        /* See if we need to do self-adjusting code.*/
        if (locemfloatstruct->adjust == FALSE) {
        	locemfloatstruct->adjust = TRUE;

        	locemfloatstruct->loops = 0;

        	/*
        	** Do an iteration of the tests.  If the elapsed time is
        	** less than minimum, increase the loop count and try
        	** again.
        	*/
        	for (; loops < CPUEMFLOATLOOPMAX; loops += loops) {
                        if (DoEmFloatIteration(abase, bbase, cbase, locemfloatstruct->arraysize, loops) > global_min_ticks) {
                                locemfloatstruct->loops = loops;
        			break;
                        }
                }
        }

        /*
        ** Verify that selft adjustment code worked.
        */
        if (locemfloatstruct->loops == 0) {
                puts("CPU:EMFPU -- CMPUEMFLOATLOOPMAX limit hit");
                free(abase);
                free(bbase);
                free(cbase);
                exit(1);
        }

        /*
        ** All's well if we get here.  Repeatedly perform floating
        ** tests until the accumulated time is greater than the
        ** # of seconds requested.
        ** Each iteration performs arraysize * 3 operations.
        */
        do {
	        total_time += DoEmFloatIteration(abase, bbase, cbase, locemfloatstruct->arraysize, locemfloatstruct->loops);
               	++iterations;
        } while (total_time < locemfloatstruct->request_secs * CLOCKS_PER_SEC);

        free(abase);
        free(bbase);
        free(cbase);

        locemfloatstruct->emflops = (double)(iterations * locemfloatstruct->loops * CLOCKS_PER_SEC) / (double)total_time;
}

/***********************
** DoEmFloatIteration **
************************
** Perform an iteration of the emulated floating-point
** benchmark.  Note that "an iteration" can involve multiple
** loops through the benchmark.
*/
static clock_t
DoEmFloatIteration(InternalFPF *abase,
                InternalFPF *bbase,
                InternalFPF *cbase,
                unsigned long arraysize, unsigned long loops)
{
        clock_t start, stop;
static unsigned char jtable[16] = {0,0,0,0,1,1,1,1,2,2,2,2,2,3,3,3};
unsigned long i;

        start = clock();

/*
** Each pass through the array performs operations in
** the followingratios:
**   4 adds, 4 subtracts, 5 multiplies, 3 divides
** (adds and subtracts being nearly the same operation)
*/
while(loops--)
{
        for(i=0;i<arraysize;i++)
                switch(jtable[i % 16])
                {
                        case 0: /* Add */
                                AddSubInternalFPF(0,abase+i,
                                  bbase+i,
                                  cbase+i);
                                break;
                        case 1: /* Subtract */
                                AddSubInternalFPF(1,abase+i,
                                  bbase+i,
                                  cbase+i);
                                break;
                        case 2: /* Multiply */
                                MultiplyInternalFPF(abase+i,
                                  bbase+i,
                                  cbase+i);
                                break;
                        case 3: /* Divide */
                                DivideInternalFPF(abase+i,
                                  bbase+i,
                                  cbase+i);
                                break;
                }
        }

        stop = clock();

        return stop - start;
}

/**************************
** SetupCPUEmFloatArrays **
***************************
** Set up the arrays that will be used in the emulated
** floating-point tests.
** This is done by loading abase and bbase elements with
** random numbers.  We use our long-to-floating point
** routine to set them up.
*/
static void
SetupCPUEmFloatArrays(InternalFPF *abase, InternalFPF *bbase, unsigned long arraysize)
{
	unsigned long i;
	InternalFPF locFPF1,locFPF2;
	/*
	** Reset random number generator so things repeat. Inserted by Uwe F. Mayer.
	*/
	randnum(13);

	for (i = 0; i < arraysize; i++) {
		Int32ToInternalFPF(randwc(50000),&locFPF1);
		Int32ToInternalFPF(randwc(50000)+1,&locFPF2);
		DivideInternalFPF(&locFPF1,&locFPF2,abase+i);
		Int32ToInternalFPF(randwc(50000)+1,&locFPF2);
		DivideInternalFPF(&locFPF1,&locFPF2,bbase+i);
	}
}